我国A股市场多因子量化选股模型实证分析

发布时间：2018-01-11 02:29

本文关键词：我国A股市场多因子量化选股模型实证分析　出处：《首都经济贸易大学》2017年硕士论文　论文类型：学位论文

【摘要】：量化投资是指通过数量化方式及计算机程序化发出买卖指令,以获取稳定收益为目的的交易方式。量化投资的研究和应用对我国投资领域具有重要意义。量化投资中的多因子选股模型是投资机构建立股票池的一种重要方法。本文针对多因子选股策略进行分析,候选因子选取纳入多维度指标数据进入模型,包括基本面类、技术面类、宏观因子类、投资者情绪类等55个指标。由于我国经济数据复杂,为解决多维度因子间的非线性相关性,本文引入了核函数来识别并提取候选因子的特征,提出核主成分分析与回归KPCR的方法,并给出了具体的实施步骤,找到了战胜市场指数的投资组合。通过具体的实验验证分析,将沪深300成分股、多时间点的指标数据作为输入的原始数据,引入核方法在高维特征空间提取了前五个主成分,其方差贡献率达到92%,比较发现其降维效果优于传统的主成分分析和偏最小二乘回归方法。将降维后的数据进行收益回归,构造出多因子选股模型,经计算该模型的MAPE值、MSE值分别为3.31%、0.00647,验证了该模型对收益具有很好的预测效果。回测发现,由该模型构造的投资组合其战胜市场基准频率为83.65%,能以大概率跑赢市场。最后得出结论,在我国股票市场数据复杂的背景下,运用核主成分回归方法可以很好地进行模式识别,找到最影响收益的显著因子,从而有效地选股建立超过市场的投资组合。
[Abstract]:Quantitative investment refers to the number of instructions issued by the sale of the way and a computer program for the purpose of the transaction to obtain stable income. The research and application of quantitative investment has important significance to our country's investment. The multi factor model to quantify the investment in stock is an important method to build investment institutions. The stock pool according to the analysis of multi factor selection strategies, selection of candidate factors into multi dimension data into the model, including the fundamentals, technical, macro factors, such as investor sentiment index. 55 because of China's economic data to solve complex nonlinear relationship between the characteristics of multi dimension factor, this paper introduces the kernel function to identify and the extraction of candidate factors, proposed method of kernel principal component analysis and regression of KPCR, and gives the specific implementation steps, found over market index portfolio. Through out Analysis and experimental verification of the CSI 300 stocks, at the same time as the original data input data, introducing the kernel extraction method of the first five principal components in high dimensional feature space, the variance contribution rate reached 92%, compared to the principal component analysis to reduce the dimension effect is better than the traditional and partial least squares regression method. The reduced dimensionality of the data of income return, construct a multi factor model of stock, by the calculation of MAPE value, MSE value were 3.31%, 0.00647, and verification of the model has a good prediction of income. Back test found that the model constructed by the portfolio over the market reference frequency 83.65%, with a large probability can outperform the market. In conclusion, the data of China's stock market complex background, using kernel principal component regression method can be good for pattern recognition, to find the most significant factor affecting revenue, thus Effective stock selection to set up portfolios beyond the market.

【学位授予单位】：首都经济贸易大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F832.51;F224

【参考文献】